| 23 Aug 2021 |
 andi- | https://github.com/NixOS/nixos-search/issues/335 | 15:22:55 |
|  collares joined the room. | 19:03:03 |
 hexa | In reply to @lukegb:zxcvbnm.ninja
update-nixos-unstable.service is wedged again How long do we wait before we explicltly mention people? | 20:39:38 |
| hexa | In reply to @lukegb:zxcvbnm.ninja
update-nixos-unstable.service is wedged again * How long do we wait before we start explicltly mentioning people? | 20:39:46 |
 Vladimír Čunát | I think we can start mentioning slowly: Amine Chikhaoui (has necessary access IIRC) | 20:53:17 |
| hexa | basically everyone on here? https://nixos.org/community/teams/infrastructure.html | 20:59:34 |
| Vladimír Čunát | Yes, that list might (still) be close to reality. | 21:10:38 |
| Vladimír Čunát | In reply to @andi:kack.it
build flags: -j2 -l2
Since when are our builders running just two cores per build? Not an issue just surprised to see this.
Actually, I wonder what kind of separation is there. I'd expect load to be shared (e.g. no VM-level separation), so load 2 is rather easy to get and you'll mostly run single-core. | 21:49:17 |
| Vladimír Čunát | In reply to @andi:kack.it
build flags: -j2 -l2
Since when are our builders running just two cores per build? Not an issue just surprised to see this.
* Actually, I wonder what kind of separation is there. I'd expect load to be shared (e.g. no VM-level separation), so load 2 is rather easy to get and you'll mostly run single-core.
It's the case where you'd want -l to be way higher than -j.
| 21:51:02 |
|  princemachiavelli joined the room. | 22:34:18 |
| princemachiavelli | Looks like /scratch is full on bastion. https://monitoring.nixos.org/prometheus/graph?g0.expr=node_filesystem_avail_bytes%7Bmountpoint%3D%22%2Fscratch%22%7D&g0.tab=0&g0.stacked=0&g0.range_input=2w | 22:35:08 |
 lukegb (he/him) | filed https://github.com/NixOS/nixpkgs/issues/135484 and mentioned the infra github team | 23:22:57 |
| 24 Aug 2021 |
 @grahamc:nixos.org | m'onit | 00:59:58 |
| andi- | In reply to @vcunat:matrix.org
Actually, I wonder what kind of separation is there. I'd expect load to be shared (e.g. no VM-level separation), so load 2 is rather easy to get and you'll mostly run single-core.
It's the case where you'd want -l to be way higher than -j.
Valid point. I wasn't thinking about that case. Perhaps worth seeing if the kernel can expose load per process namespace? | 09:46:33 |
| Vladimír Čunát | In reply to @andi:kack.it
Valid point. I wasn't thinking about that case. Perhaps worth seeing if the kernel can expose load per process namespace? And then we patch all the build systems? I'm more hopeful for a job-server... or us separating -l from -j. | 09:49:36 |
| andi- | No, I would patch the way the nix sandbox works. | 09:50:43 |
| andi- | e.g. if you don't run enough work for two cores your load should not be >=2 | 09:51:04 |
| Vladimír Čunát | I didn't think that linux namespaces allow modifying the load, but perhaps... I haven't given it much time. | 11:25:32 |
| Vladimír Čunát | * I didn't think that linux namespaces allow modifying/unsharing the load, but perhaps... I haven't given it much time. | 11:25:50 |
| andi- | In reply to @vcunat:matrix.org
I didn't think that linux namespaces allow modifying/unsharing the load, but perhaps... I haven't given it much time. Me neither but that would seem like a reasonable thing for me. Of course the actual schedulers might not make this approachable. | 11:38:06 |
| lukegb (he/him) | Oh, lxc emulates the loadavg in lxcfs by using the cgroups cpuacct data. Huh. | 19:33:20 |
| lukegb (he/him) | Anyway, this is more a Nix dev question, no? | 19:33:46 |
| andi- | In reply to @lukegb:zxcvbnm.ninja
Anyway, this is more a Nix dev question, no? perhaps it devolved in that direction, yes. | 22:09:10 |
| andi- | I posted the gist of the discussion so far over there. Perhaps we can continue there. | 22:12:05 |
|  aanderse joined the room. | 22:34:07 |
| 25 Aug 2021 |
|  baloo joined the room. | 06:02:24 |
| hexa | could I get restart bits on hydra, please? | 13:23:59 |
| hexa | * could I get restart bits on hydra, please? it would be helpful because I'm taking care of staging runs here and there. | 13:27:05 |
 Amine Chikhaoui | eh just realized that my IP is stale for bastion, grahamc (he/him) or Rob Vermaas (he/him) maybe someone can update it to 50.116.57.221. I moved my linode server to another region a while back. | 17:03:41 |
| Amine Chikhaoui | but not urgent :) I was about to take care of hexa request and realized that | 17:04:14 |
